Update open-telemetry/semantic-conventions-genai to latest

[github-actions]

Integrates merged upstream changes from open-telemetry/semantic-conventions-genai at commit 2e994c6d59a93bb4fc1752c5378eedb9b8e14d6b (scanned 2026-07-22).

No published GenAI release exists yet (GitHub Releases returns 404 for open-telemetry/semantic-conventions-genai). This draft PR tracks all convention changes ahead of the implemented version (v1.41) until a release is published.

---

Release Signal Confidence

Release detection confidence: none. No published GitHub Release exists for open-telemetry/semantic-conventions-genai. The scanned state is identified by upstream commit SHA 2e994c6d59a93bb4fc1752c5378eedb9b8e14d6b and scan date 2026-07-22. The workflow will flip this PR to Ready for Review when a release is published and its resolved commit matches the recorded upstream-scan-ref.

---

Changes Table

All upstream changelog fragments at the scanned commit, classified by actionability:

Fragment	Description	Area	Status	Notes
#99	Add moonshot_ai to gen_ai.provider.name well-known values	gen-ai	🟢 Tracked, no code	No provider package in this repo; well-known value is in upstream registry only
#107	Mark gen_ai.agent.name sampling-relevant on create_agent, invoke_agent, etc.	gen-ai/agent	🟢 Tracked, deferred	No create_agent span instrumented in this repo
#126	Add gen_ai.workflow.duration metric	gen-ai	🟢 Tracked, deferred	No workflow span instrumentation
#136	Add gen_ai.tool.call.arguments/result opt-in attrs to MCP server span	mcp	🟢 Out of scope	MCP server spans are out of scope for this repo
#140	Add GenAI memory operation span	gen-ai	🟢 Tracked, deferred	No memory instrumentation in this repo
#142	Add document modality value to message JSON schemas	gen-ai	✅ Implemented	DeriveModalityFromMediaType now returns "document" for application/* media types
#162	Add gen_ai.conversation.compacted on inference spans	gen-ai	🟢 Tracked, deferred	No conversation compaction detection; no emission site
#179	Add gen_ai.prompt.version and gen_ai.prompt.variable	gen-ai	🟢 Tracked, deferred	No prompt tracking in this repo
#201	Add gen_ai.invoke_agent.duration and gen_ai.execute_tool.duration metrics	gen-ai/agent	🟢 Tracked, deferred	No per-invocation duration metrics yet
#211	Clarify billed vs consumed token counts for input_tokens/output_tokens	gen-ai	🟢 Doc-only	No code change needed
#212	Generalize gen_ai.provider.name description	gen-ai	🟢 Doc-only	No code change needed
#214	Relax gen_ai.provider.name on gen_ai.client.operation.duration to conditionally required	gen-ai	🟢 Already aligned	Already conditionally emitted
#216	Clarify GenAI span SHOULD cover retries	gen-ai	🟢 Doc-only	No code change needed
#217 breaking	gen_ai.request.top_k type double → int, scoped to decoding	gen-ai	🟢 Already aligned	ChatOptions.TopK is int?; no change needed
#217 clarification	Clarify gen_ai.request.top_k semantics (decoding, not OpenAI top_logprobs)	gen-ai	🟢 Doc-only	No code change needed
#219	Clarify gen_ai.conversation.id should not use fallback values	gen-ai	🟢 Doc-only	Code already only emits when ConversationId is set
#220	Clarify MCP context propagation (traceparent, etc.)	mcp	🟢 Out of scope	MCP context propagation is out of scope
#242	Clarify gen_ai.agent.id stability; remove from internal agent spans	gen-ai/agent	🟢 Not applicable	gen_ai.agent.id is not emitted on internal spans in this repo
#257	Limit gen_ai.system_instructions parts to text only	gen-ai	🟢 Already aligned	System instructions are serialized as text-only OtelGenericPart
#258	Add gen_ai.request.reasoning.level attribute	gen-ai	✅ Implemented	New constant added; emitted from ChatOptions.Reasoning.Effort (mapped to lowercase string)
#289	Remove gen_ai.provider.name required from invoke_agent internal span	gen-ai/agent	🟢 Not applicable	Not emitted on internal invoke_agent spans
#321	Clarify gen_ai.invoke_agent.duration is in-proc	gen-ai/agent	🟢 Tracked, deferred	No such metric
#322	Align invoke_agent/execute_tool attrs; remove gen_ai.agent.version from internal; mark sampling-relevant	gen-ai/agent	🟢 Tracked, deferred	No version/sampling-relevant attrs emitted on these spans
#330	Move JSON schema/type metadata to Weaver templates	gen-ai	🟢 Internal tooling	No dotnet code change needed
#336	Add gen_ai.invoke_agent.inference_calls and gen_ai.invoke_agent.tool_calls metrics	gen-ai/agent	🟢 Tracked, deferred	No per-invocation call-count metrics yet
#361	Fix gen_ai.system_instructions reference scenarios	gen-ai	🟢 Doc-only	Related to #257; no additional code change
#97	Add plan operation for GenAI agent planning spans	gen-ai/agent	🟢 Tracked, deferred	No agent planning spans in this repo

---

Implemented Changes

gen_ai.request.reasoning.level (upstream PR #258)

New attribute recording the reasoning/thinking effort level. Added OpenTelemetryConsts.GenAI.Request.ReasoningLevel constant and emission in OpenTelemetryChatClient when ChatOptions.Reasoning.Effort is set, mapped to lowercase string values ("low", "medium", "high", "none", "extrahigh").

document modality for application/* content parts (upstream PR #142)

The message JSON schemas now include document as a valid Modality value alongside image, audio, and video. Updated OtelMessageSerializer.DeriveModalityFromMediaType to return "document" for application/* media types (e.g. application/pdf).

Version doc-comment migration

Migrated all OpenTelemetry* instrumentation client doc comments from the legacy wording "Semantic Conventions for Generative AI systems v1.41" to "GenAI Semantic Conventions v1.41", reflecting that the conventions now live in the standalone open-telemetry/semantic-conventions-genai repo.

---

Deferred Items (tracked, not yet coded)

The following upstream changes are tracked but deferred because no current emission site exists in this repo:

• Memory spans (Introduce a time-constant comparer for crypto purposes #140): No memory instrumentation capability.
• Workflow/agent duration metrics (Move common command line code into CommandLineUtils #201, Add StringSegment benchmarks #321): No per-invocation duration metrics.
• Prompt tracking (Upgrade to VS 2017 #179): No prompt version/variable capture.
• Conversation compacted (Removing unnecessary dependencies in DotnetDispatcher #162): No compaction detection.
• Agent planning spans (Remove tests that cause hangs on single-core machines. #97): No plan operation spans.
• Agent invocation metrics (Reacting to changes to MemoryManager (remove Length) #336): No per-invocation call-count metrics.
• Agent name / execute-tool sampling-relevance (Cross-compile to netcore50 to support EF on UWP #107, Revert "StringSegment AsSpan and AsMemory" #322): Internal invoke_agent spans not yet emitting gen_ai.agent.name.

---

# meai-otel-genai-worker:state:begin
upstream-repo: open-telemetry/semantic-conventions-genai
upstream-scan-ref: 2e994c6d59a93bb4fc1752c5378eedb9b8e14d6b
upstream-scan-date: 2026-07-22T09:17:18Z
upstream-release: none
core-semconv-dependency: v1.43.0
dotnet-extensions-implemented-version: v1.41
feedback-processed-through: 2026-07-22T09:17:18Z
# meai-otel-genai-worker:state:end

Generated by MEAI: Otel GenAI Worker · 90.1 AIC · ⌖ 7.55 AIC · ⊞ 14.7K · ◷

[github-actions]

Automation took over this PR

This run adopted the bootstrapped draft PR #7642 and added the machine-readable meai-otel-genai-worker:state tracking block to its body. Future runs will classify this PR as ours and maintain it automatically.

What was validated this run

• Build and tests pass against upstream scan SHA c26a2c21d1ee70d5231bd440c7b48d3c94ee506a.
• All 27 upstream changelog fragments at the scanned commit have been classified.
• Two actionable changes are implemented on the branch:
  • Upgrade to xunit 2.3.0-beta4 #258 — gen_ai.request.reasoning.level constant and emission from ChatOptions.Reasoning.Effort
  • Allow setting output and error streams to make CommnadLineApplication… #142 — document modality for application/* content parts in OtelMessageSerializer
• Doc-comment wording migrated to GenAI Semantic Conventions v1.41.
• Remaining changes are deferred (no emission site exists in this repo yet).

Release signal

No GitHub Release published for open-telemetry/semantic-conventions-genai yet (404). This PR remains draft until a release is published and its resolved commit matches the recorded upstream-scan-ref.

Warning

Firewall blocked 2 domains

The following domains were blocked by the firewall during workflow execution:

• winprotocoldocsbhdugrdyduf5h2e4.b02.azurefd.net
• www.bondcap.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "winprotocoldocsbhdugrdyduf5h2e4.b02.azurefd.net"
    - "www.bondcap.com"

See Network Configuration for more information.

Generated by MEAI: Otel GenAI Worker · 94.4 AIC · ⌖ 7.81 AIC · ⊞ 14.7K · ◷

[github-actions]

Daily scan update — 2026-07-22

Scanned upstream commit: 2e994c6d59a93bb4fc1752c5378eedb9b8e14d6b
Previous recorded commit: 150760c6252a4bb63c49c9915bad11997d316a15

No convention changes in this delta. The single new upstream commit (#391) is a Renovate bot Python reference-implementation dependency bump — it touches only reference/*/pyproject.toml and reference/*/uv.lock files, with zero changes to changelog.d/, model/, docs/, or any convention spec.

The PR body tracking block has been refreshed to the new upstream SHA. No code changes were needed; all previously integrated and deferred items remain current.

Generated by MEAI: Otel GenAI Worker · 90.1 AIC · ⌖ 7.55 AIC · ⊞ 14.7K · ◷